In the realm of AI infrastructure, there’s a noticeable trend where popular open-source tools transition into venture-backed companies that accumulate value in the hundreds of millions. A recent case in point is RadixArk, which focuses on developing SGLang, a tool that enables faster and more cost-effective AI model execution.
RadixArk has been valued at roughly $400 million during a funding round led by Accel, as reported by sources familiar with the situation. This valuation is impressive given that the startup was only founded last August, though TechCrunch could not confirm the exact funding details.
This update is linked to a team that has been working on SGLang—used by firms like xAI and Cursor for enhancing the speed of AI model training—who have now joined this newly formed commercial startup. RadixArk originated as SGLang in the lab of Ion Stoica, a co-founder of Databricks, at the University of California, Berkeley in 2023.
The startup had earlier attracted angel investment from notable figures, including Intel CEO Lip-Bu Tan, according to the sources.
Ying Sheng, a prominent developer of SGLang and a former engineer at xAI, has reportedly left to take on the role of co-founder and CEO at RadixArk. She announced her position on LinkedIn last month and previously held a research role at Databricks.
Attempts to reach comments from RadixArk’s Ying Sheng, Accel, and Lip-Bu Tan were unsuccessful.
Both SGLang and RadixArk aim to enhance inference processing efficiency, which allows models to operate more effectively without requiring different hardware. Since inference and model training typically constitute significant server costs for AI services, utilizing these optimization tools can lead to substantial immediate savings.
RadixArk is not the only one making this transition from an open-source initiative to a funded startup. A similar project, vLLM, has also achieved notable progress in optimizing inference. Reports suggest that this new company is negotiating to raise over $160 million, aiming for a valuation around $1 billion, according to Forbes last month.
Andreessen Horowitz is reportedly leading the investment into vLLM, though specifics of the funding remain undisclosed as the firm declined to comment. Interestingly, vLLM’s co-founder Simon Moe described reports on the investment round as “factually inaccurate,” without detailing which aspects were misleading.
Like SGLang, vLLM has its roots in Ion Stoica’s lab at UC Berkeley. Stoica is renowned as a co-founder of Databricks and several other startups.
Brittany Walker, a general partner at CRV, highlighted that many major tech firms now utilize vLLM for inference tasks, while SGLang has seen a surge in popularity over the last six months, although her firm has not invested in either.
RadixArk continues to enhance SGLang as an open-source engine for AI models and is also developing Miles, a framework for reinforcement learning that enables companies to train AI models to improve progressively.
While much of the company’s offerings remain free, it appears RadixArk has begun charging for certain hosting services, as noted by sources.
In recent months, startups providing inference infrastructure for developers have experienced a funding boom, which highlights the ongoing significance of the inference layer in AI. For instance, Baseten landed $300 million that valued the company at $5 billion, as The Wall Street Journal noted. Similarly, its rival, Fireworks AI, also saw significant funding, reaching a $4 billion valuation last October after raising $250 million.
